- C# 99.6%
- Shell 0.4%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
GoalSpec.SensoryGain (default 1, in [0,1]) is held in DriveState.SensoryGain, written only by the Actor at takeover (at construction for the Start goal) and handed to Perception.Sense. The gain multiplies each channel's afferent as it leaves the conduction ring, so sensitisation still follows the world; Load is built from the gated afferents; ChannelStimulus stays the world's; the nociceptive region afferents, the withdrawal reflex and its flinch are not gated. lodur:ruling:stress-bar-shows-arousal-calm-silences Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> |
||
| docs | ||
| src | ||
| tests | ||
| tools/pain-box-golden | ||
| .gitattributes | ||
| .gitignore | ||
| Directory.Build.props | ||
| Lodur.slnx | ||
| README.md | ||
Lodur
A physics character middleware for characters that feel and react: they pull back from something hot, jump when poked, and, if they want their objective badly enough, crawl toward it after losing a leg.
Status: layers 1-3 and a layer-4 stand-in run on a PD-muscle ragdoll on Ymir. The
design is docs/lodur-target.md; read it first. Lodur.Core
holds perception, the withdrawal reflex, the drive state and a hold-against-reflex
stand-in, run per character by Actor; Lodur.Body.Ymir is the PD-muscle active ragdoll
on Ymir's Box3D SoA world (docs/first-slice.md, measured and
unmeasured). Behaviour selection beyond the stand-in, impairment and the learned tracker
are not coded.
Entry points: src/Lodur.Core/Actor.cs (one character's nervous state, one tick per
physics step), src/Lodur.Core/PdMuscle.cs (engine-free PD law, strength per muscle),
src/Lodur.Body.Ymir/HumanoidRagdoll.cs (the ragdoll, Drive and ReadRegions),
tests/Lodur.Core.Tests (the layers and parity with the pain box; builds without Ymir),
tests/Lodur.Tests (stand, fall, rest, same-seed-same-hash, and the connected pain box:
ConnectedPainBoxTests). The map this was cut from is
docs/connect-rungs-map.md. Needs a sibling Ymir checkout at
main (the SoA joints merge, 2026-11-10, or later); dotnet test tests/Lodur.Tests builds
its native library with cmake.
The problem
Active-ragdoll middleware (NaturalMotion's Euphoria, RootMotion's PuppetMaster) and learned physics controllers (DReCon, ProtoMotions) solve the body: physics, pose tracking and, in the learned case, balance. None of them models the nervous system. They have no senses that are not forces, no reflexes layered under deliberate behaviour, and no character whose pain competes with its will over time. Lodur is that layer.
Shape (planned)
- Core (C#, no engine references): perception, reflexes, a drive state of pain, shock, adrenaline and resolve that evolves every tick (the will breaks at zero resolve and mends as it recovers, and can break again), and behaviour selection that turns the drive into targets plus muscle strength, weighted by the will available to each voluntary pose.
- Body (per engine): follows those targets on the engine's own articulated physics. v1 is a PD-tracking active ragdoll on Ymir; v2 adds a small learned tracker for upright balance behind the same interface.
First consumer
Bad Day Simulator, a 3D physics sandbox.
Where it lives
Canonical: https://eureka.gamecult.org/forge/GameCult/Lodur. GitHub
GameCult/Lodur is a push mirror.
Prior-art research:
gamecult-ops/docs/research/euphoria-at-home-prior-art.md.